Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis

WORKING PAPER 19
Crop Production and Road
Connectivity in Sub-Saharan
Africa: A Spatial Analysis
Paul Dorosh, Hyoung-Gun Wang, Liang You,
and Emily Schmidt
FEBRUARY 2009

© 2009 The International Bank for Reconstruction and Development / The World Bank
1818 H Street, NW
Washington, DC 20433 USA
Telephone: 202-473-1000
Internet: www.worldbank.org
E-mail: feedback@worldbank.org
All rights reserved
A publication of the World Bank.
The World Bank
1818 H Sreet, NW
Washington, DC 20433 USA
The findings, interpretations, and conclusions expressed herein are those of the author(s) and do not necessarily reflect the views of the
Executive Directors of the International Bank for Reconstruction and Development / The World Bank or the governments they represent.
The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other
information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory
or the endorsement or acceptance of such boundaries.
Rights and permissions
The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of
applicable law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will
normally grant permission to reproduce portions of the work promptly.
For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance
Center Inc., 222 Rosewood Drive, Danvers, MA 01923 USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com.
All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818
H Street, NW, Washington, DC 20433 USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org.

About AICD
This study is a product of the Africa Infrastructure Country
Diagnostic (AICD), a project designed to expand the
world’s knowledge of physical infrastructure in Africa.
AICD will provide a baseline against which future
improvements in infrastructure services can be measured,
making it possible to monitor the results achieved from
donor support. It should also provide a better empirical
foundation for prioritizing investments and designing
policy reforms in Africa’s infrastructure sectors.
AICD is based on an unprecedented effort to collect
detailed economic and technical data on African
infrastructure. The project has produced a series of reports
(such as this one) on public expenditure, spending needs,
and sector performance in each of the main infrastructure
sectors—energy, information and communication
technologies, irrigation, transport, and water and sanitation.
Africa’s Infrastructure—A Time for Transformation,
published by the World Bank in November 2009,
synthesizes the most significant findings of those reports.
AICD was commissioned by the Infrastructure Consortium
for Africa after the 2005 G-8 summit at Gleneagles, which
recognized the importance of scaling up donor finance for
infrastructure in support of Africa’s development.
The first phase of AICD focused on 24 countries that
together account for 85 percent of the gross domestic
product, population, and infrastructure aid flows of Sub-
Saharan Africa. The countries are: Benin, Burkina Faso,
Cape Verde, Cameroon, Chad, Côte d'Ivoire, the
Democratic Republic of Congo, Ethiopia, Ghana, Kenya,
Lesotho, Madagascar, Malawi, Mozambique, Namibia,
Niger, Nigeria, Rwanda, Senegal, South Africa, Sudan,
Tanzania, Uganda, and Zambia. Under a second phase of
the project, coverage is expanding to include as many other
African countries as possible.
Consistent with the genesis of the project, the main focus is
on the 48 countries south of the Sahara that face the most
severe infrastructure challenges. Some components of the
study also cover North African countries so as to provide a
broader point of reference. Unless otherwise stated,

therefore, the term “Africa” will be used throughout this
report as a shorthand for “Sub-Saharan Africa.”
The World Bank is implementing AICD with the guidance
of a steering committee that represents the African Union,
the New Partnership for Africa’s Development (NEPAD),
Africa’s regional economic communities, the African
Development Bank, the Development Bank of Southern
Africa, and major infrastructure donors.
Financing for AICD is provided by a multidonor trust fund
to which the main contributors are the U.K.’s Department
for International Development, the Public Private
Infrastructure Advisory Facility, Agence Française de
Développement, the European Commission, and Germany’s
KfW Entwicklungsbank. The Sub-Saharan Africa Transport
Policy Program and the Water and Sanitation Program
provided technical support on data collection and analysis
pertaining to their respective sectors. A group of
distinguished peer reviewers from policy-making and
academic circles in Africa and beyond reviewed all of the
major outputs of the study to ensure the technical quality of
the work.
The data underlying AICD’s reports, as well as the reports
themselves, are available to the public through an
interactive Web site, www.infrastructureafrica.org, that
allows users to download customized data reports and
perform various simulations. Inquiries concerning the
availability of data sets should be directed to the editors at
the World Bank in Washington, DC.

iii
Contents
Abstract iv
Abbreviations and acronyms v
Acknowledgments v
1 Data 2
Agroecological zones 2
IFPRI SPAM 2
Transport networks and travel times 3
Measures of local market size 4
Roughness of terrain 4
Merging of data sets 5
2 Descriptive statistics 5
Crop production 5
Travel time, population, and agricultural production 6
Estimated crop-surplus regions 8
3 Conceptual framework and model 12
Econometric issues 13
4 Econometric estimates 14
The impacts of road connectivity on crop production 14
Road connectivity and adoption of high-input/high-yield technology: Sub-Saharan Africa total 16
Comparing East and West Africa 17
5 Interpretation of results and implications for spatial investment 24
Implications of investment: corridors versus rural roads in Mozambique 26
6. Summary and policy implications 30
References 32
Appendixes 34

iv
Abstract
This study adopts a cross-sectional spatial approach to examine the impact of transport infrastructure
on agriculture in Sub-Saharan Africa using new data obtained from geographic information systems
(GIS). Our approach involves descriptive statistical analysis and econometric regressions of crop
production or choice of technology for each location (a 9x9 kilometer pixel) in Sub-Saharan Africa on (a)
agroecological zones and crop production potentials by the Food and Agriculture Organization (FAO) and
the International Institute for Applied Systems Analysis (IIASA), (b) GIS data on crop production from
the International Food Policy Research Institute’s (IFPRI) spatial crop allocation model (SPAM), and (c)
road infrastructure based largely on data from the United Nations Environment Programme (UNEP) and
estimated travel times.
We address three main issues. First, we analyze the impact of road connectivity on crop production
and choice of technology when we control basic supply and demand factors. Second, we investigate the
impact on agricultural output of investments that reduce travel time on roads of various types. Third, we
provide an example of how this type of analysis could be used to construct benefit-cost ratios of
alternative road investments in terms of enhanced agricultural output per dollar invested.
We find that agricultural production and proximity (as measured by travel time) to urban markets are
highly correlated, even after taking agroecology into account. Likewise, adoption of high-
productive/high-input technology is negatively correlated with travel time to urban centers.
There is substantial scope for increasing agricultural production in Sub-Saharan Africa, particularly in
more remote areas. Total crop production relative to potential production is 45 percent for areas within
four hours’ travel time from a city of 100,000 people. In contrast, it is just 5 percent for areas more than
eight hours away. These differences in actual versus potential production reflect the relatively small share
of land cultivated out of total arable land in more remote areas.
For remote regions, low population densities and long travel times to urban centers sharply constrain
production. Reducing transport costs (travel time) to these areas would expand the feasible market size for
these regions, easing the constraint on production. If the expansion in production from these areas were
small in terms of the relevant regional, national, or subnational market, average market prices outside the
formerly remote region would be unaffected, and significant aggregate production increases could result.
We find some interesting differences between East Africa and West Africa. On average, East Africa
has lower population density, smaller local markets, and lower road connectivity—the average travel time
to the nearest city is more than twice that in West Africa. While average suitable area for crop production
is similar in East and West Africa, average crop production per pixel in East Africa is just 30 percent of
that in West Africa. Road connectivity has different impacts in the two regions. In East Africa the results
are similar as for all Sub-Saharan Africa. Longer travel time decreases total crop production, and reducing
travel time significantly increases adoption of high-input/high-yield technology in East Africa, but the
impacts are insignificant for West Africa. This may be because the more densely roads are connected, the
smaller the marginal benefits of more connections. West Africa already has a relatively well-connected
road network.

v
Abbreviations and acronyms
AEZ agroecological zone
CIESIN Center for International Earth Science Information Network
ESRI Environmental Systems Research Institute
FAO Food and Agriculture Organization
FAOSTAT Food and Agriculture Organization Statistical Databases
GDP gross domestic product
GIS geographic information system
GPW–UR Gridded Population of the World, version 3, with Urban Reallocation
GRUMP Global Rural-Urban Mapping Project
IFPRI International Food Policy Research Institute
IIASA International Institute for Applied Systems Analysis
kg/ha kilogram per hectare
km/hr kilometer per hour
LUTs land utilization types
ORNL Oak Ridge National Laboratory
SI crop suitability index
SPAM spatial crop allocation model
SRTM Shuttle Radar Topography Mission
UNEP United Nations Environment Programme
Acknowledgments
We wish to thank Stanley Wood of IFPRI for sharing the Spatial Allocation Model output and for his
helpful comments and suggestions throughout this process. Siobhan Murray from DECRG was
instrumental in the success of this research; her support with data and continual guidance were invaluable.
We would also like to thank Vivien Foster (AFTSN) and Marisela Montoliu Munoz (FEU) for this
research support, and the Spatial and Local Development Team and external participants for valuable
feedback during the World Bank seminar in 2008 where we presented preliminary results of the analysis.
About the authors
Paul Dorosh, Hyoung-Gun Wang, and Emily Schmidt are economists at the World Bank. Liang You
is an economist with the International Food Policy Research Institute. The corresponding author is
Hyoung-Gun Wang, hwang4@worldbank.org.

here is substantial evidence that investments in roads and road connectivity positively effect
agricultural productivity and output. Such evidence includes econometric analysis of subnational
data on the effects of public spending (roads, agricultural research, education, and so on) on
agricultural output, incomes, and poverty in the People’s Republic of China and India (Fan and Hazell
2001). Econometric analysis of household data on the effects of road connectivity on input use, crop
output, and household incomes in Madagascar and Ethiopia (Chamberlin and others 2007; Stifel and
Minten 2008) suggests that remoteness negatively affects agricultural productivity and incomes at the
household level. The impacts of road infrastructure on agricultural output and productivity are particularly
important in Sub-Saharan Africa for three reasons. First, the agricultural sector accounts for a large share
of gross domestic product (GDP) in most Sub-Saharan countries. Second, poverty is concentrated in rural
areas. Finally, the relatively low levels of road infrastructure and long average travel times result in high
transaction costs for sales of agricultural inputs and outputs, and this limits agricultural productivity and
growth. Thus, investments in road infrastructure can have a significant impact on rural and national
incomes through their effects on agriculture.
Lack of detailed household and subnational data, particularly time-series data, preclude analysis of
the impacts of transport infrastructure on agriculture for most of Sub-Saharan Africa. Instead, this study
adopts a cross-sectional spatial approach to examine the impact of transport infrastructure on agriculture
in Sub-Saharan Africa using newly developed geographic information system (GIS) data on (a)
agroecological zones and crop production potentials by the food and agriculture organization (FAO) and
the International Institute for Applied Systems Analysis (IIASA), (b) GIS data on crop production from
the International Food Policy Research Institute (IFPRI) spatial crop allocation model (SPAM), and (c)
road infrastructure based largely on United Nations Environment Programme (UNEP) data and estimated
travel times (Thomas 2007). Our approach involves econometric regressions of crop production or choice
of technology for each location as a function of crop production potentials (which are in turn determined
by agroecology and agronomic characteristics of individual crops), travel time, and other factors. But, as
discussed below, econometric estimation using this cross-section spatial data presents several challenges,
including controlling for endogeneity of road placement (construction), possible omitted variables
(unobserved fixed effects), and measurement errors in estimation. Moreover, these reduced-form
equations of agricultural production or input across locations do not—in the context of relatively
abundant supply of land—readily capture constraints on overall market demand even with the inclusion of
variables that proxy availability of markets. For this reason, we place these econometric results in a
broader analytical framework that explicitly incorporates demand constraints. We address three main
issues. First, we analyze the impact of road connectivity on crop production and choice of technology
when we control basic supply and demand factors. Second, we investigate the impact on agricultural
output of investments that reduce travel time on roads of various types. Third, we provide an example of
how this type of analysis could be used to construct benefit-cost ratios of alternative road investments in
terms of enhanced agricultural output per dollar invested.
T

CROP PRODUCTION AND ROAD CONNECTIVITY IN SUB-SAHARAN AFRICA
2
This paper is organized as follows. Section 1 describes the databases used and presents basic
descriptive statistics. Section 2 discusses the econometric methodology. Section 3 presents results of
regressions of agricultural output and choice of technology for various Sub-Saharan crops and regions.
Section 4 presents results of a simulation of improvements in road quality. The final sections, 5 and 6,
summarize the paper, present policy implications, and discuss possible further work.
1 Data
Three major types of spatial data are used in the analysis: agroecological zones and crop potentials as
derived by FAO and IIASA, estimates of the spatial allocation of crops from the IFPRI SPAM, and
various measures of economic distance constructed on the basis of estimated travel times and market size.
This section presents a brief description of this data and details of the SPAM.
Agroecological zones
The agroecological zone (AEZ) methodology developed by FAO and the IIASA combines
georeferenced data on land resources (climate, soil, and terrain) with a mathematical model for the
calculation of potential biomass and yields per crop and management system. Land resource data include
an adjustment for estimated land requirements for housing and infrastructure, based on population and
population density. The FAO-IIASA model produces biomass and yield estimates for 154 different land
utilization types (LUTs) comprising various rainfed and irrigated crops, and fodder and grassland land
uses, each at three generic levels of inputs and management (high, intermediate, and low).1
The resolution
of the model is 0.5-degree latitude/longitude cells. Crop agronomic potential in each location is
summarized in the Crop Suitability Index (SI), which is defined as: SI = VS*0.9 + S*0.7 + MS*0.5 +
NS*0.3, where VS, S, MS, and NS denote percentages of the grid cell with attainable yields that are 80
percent or more, 60–80 percent, 40–60 percent, and 20–40 percent of maximum potential yield. SI is
essentially a measure of quality adjusted land area (the adjusted share of land suitable for cultivation of a
particular crop or group of crops), with a maximum value of 0.9 and a minimum value of 0.2
IFPRI SPAM
The IFPRI SPAM (You and others 2007b) is designed to estimate the spatial distribution of crop
production in Sub-Saharan Africa. The model combines data on national and subnational crop area and
production (three-year average, 1999–2001), land cover, and land suitability using a cross-entropy
regression to estimate the spatial allocation of crops at a resolution of 5x5 minutes (approximately 9x9
kilometers on the equator).3
Specifically, the model takes as inputs (a) crop area (the total physical area cultivated at least once per
year from subnational estimates, generally from official government statistics), (b) total crop land (from
1
For irrigated land, only high and intermediate levels of inputs are defined.
2
For more information, consult FAO (2003, 1981); FAO, IFPRI, SAGE (2006); and Fischer and others (2001).
3
The number of pixels per country is given in table A.1.

3
Africa Land Cover 2000), (c) suitable area (extracted from the FAO/IIASA global crop suitability
surfaces), and (d) irrigated land area (from the global map of irrigation, Siebert and others 2001).
The model algorithm begins with an initial estimate of crop area and production by location derived
from available production statistics and corresponding biophysical suitability maps. It then uses a cross-
entropy regression to create a new allocation of crop area and production that satisfies a set of constraints
yet minimizes the differences between the initial and final allocations. These constraints are:
• The sum of allocated crop areas/production is equal to existing statistical data.
• The actual agricultural area from the satellite image is the upper limit.
• The sum of allocated crop area cannot exceed the suitable area for the particular crop.
• The sum of allocated irrigated areas cannot exceed the area in the Africa map of irrigation.
The model produces estimates for four production systems: rain-fed/high-input, rain-fed/low-input,
irrigated, and subsistence4
for each of the following 20 crops:
• Six cereals: wheat, rice, maize, barley, millet, sorghum
• Seven pulses: potato, sweet potato and yam, cassava, plantain and banana, soybean, beans, other
pulses
• Five fibers: sugarcane, sugar beets, coffee, cotton, other fibers
• Two others: groundnuts, other oil crops
In section 5, we present regression results for crop production for four crop groups: maize, cereals
(wheat, rice, maize, barley, millet, and sorghum), cash crops (coffee, cotton, groundnuts), and the total of
the 20 crops. In all cases, crop production is measured in U.S. dollars, using the median of the dollar price
crop indices across Sub-Saharan Africa (see table A.2). Each crop or crop group is also disaggregated into
three different production systems (rain-fed/high-input, rain-fed/low-input, including subsistence, and
irrigated).
Transport networks and travel times
We constructed several road-connectivity measures, including distance to roads and travel time. We
used UNEP road data5
to calculate distance to the nearest type-1 (motorway/major road), type-2 (all
weather/improved), and type-3 (partially improved/earth) roads.6
For travel-time measures, we used
4
In our analysis presented below, we combine low-input/rain-fed and subsistence systems together, since they share
similar characteristics and because estimates of the location of subsistence crop areas were generated based on the
distribution of population using assumptions of low-input/rain-fed technology. See You and others (2007a, 2006).
5
The United Nations Environment Programme (UNEP) road map was created as part of the fourth version of the
Africa Population Database by Andy Nelson. The input road maps were from Digital Chart of the World
(Environmental Systems Research Institute, ESRI, 1993) and Michelin Travel Publications (2004, Michelin Travel
Assistance Services).
6
Using spatial analysis tools in ESRI ArcGIS software, we calculate travel time from each point (pixel) to the
nearest city of a specified size in Sub-Saharan Africa. These travel time raster data are exported to a standard-format
database and then merged with the SPAM data sets for analysis. Note, however, that these estimates depend greatly

4
UNEP road data supplemented with national road information for Ethiopia, Uganda, and Kenya to
construct separate variables for travel time to towns or cities of (a) 25,000 people or more, (b) 100,000
people or more, and (c) 500,000 people or more (Thomas 2007).7
Total length and the estimated travel
speed of each road type are shown in table A.3.
The road data are geoprocessed to create friction grids, and a one-hour delay is added in crossing
country borders (see table A.8). To identify the nearest city and its population size, we used Global Rural-
Urban Mapping Project (GRUMP) population data from the Center for International Earth Science
Information Network (CIESIN).8
These population counts for the year 2000 were adjusted to match UN
totals. We then combined friction grids and the locations of cities with different sizes and calculated
travel time to the nearest town or city of (a) 25,000 people or more, (b) 100,000 people or more, and (c)
500,000 people or more (Thomas 2007).9
Measures of local market size
Local market size, in addition to accessibility (travel time), may also influence crop production. There
is no consensus on defining the boundary or size of a local market (or market potential measure), but a
standard method is to use a distance-decay model and calculate population aggregates decayed over
distance.10
Thus, local market size, i, is calculated as:
local market sizei ik k
k
w pop=
where popk is the population aggregate in neighboring area k and
the distance weight wk,j =1/(dk,i) and
where dk,i is the Euclidean distance between k and i in kilometers and is the decay parameter.
The choice of the decay parameter is arbitrary. In this regard, we experimented with various measures
of distance-weighted population aggregates and came up with two proxy variables: (a) a population count
in its own pixel and (b) a distance-weighted population aggregate in neighboring areas (excluding its own
population). We define neighboring areas as the areas within a 100-km radius. We divide these areas into
six subgroups (radius between 1–2 km, 2–5 km, 5–10 km, 10–20 km, 20–50 km, and 50–100 km), as
listed in table A.4. The input data from the GRUMP population counts in year 2000, at a 1-km resolution.
Roughness of terrain
To control for the endogeneity of road construction, we create a terrain-roughness variable.
Specifically, we use the Shuttle Radar Topography Mission (SRTM) version 3, terrain elevation grids,
on quality of the input geographic information system (GIS) data on road networks, which need to be updated
frequently. See Murray (2007) for a description of the data.
7
These calculations assume travel speeds of 50, 35, and 25 km/hr for type-1, type-2, and type-3 roads, respectively.
8
Specifically, it is the Gridded Population of the World, version 3, with Urban Reallocation (GPW-UR).
9
Details of the calculations for Mozambique are given in Dorosh and Schmidt (2008).
10
See Deichmann (1997) for a review of the issues related to this methodology.

5
subtract the minimum from the maximum for each 30x30 arc second cells (approximately 1x1 km) in a
pixel, and define it as the terrain roughness.
Merging of data sets
Data sets of different resolutions (grid sizes) are aggregated at the IFPRI SPAM data resolution (5x5
minutes, approximately 9x9 km on the equator), exported to standard-format data sets, and then merged
with the SPAM database. In aggregation, we control for differences in land area (mainly due to oceans or
lakes), such that:
( )
=
i
i
i
SPAMk
k
SPAMk
kk
SPAM
area
areatimetravelold
timetravelnew
__
__
where areak is land area (km2
) at a 1-km resolution from GRUMP (CIESIN). The final data set includes
42 countries, which are listed in table A.1.
2 Descriptive statistics
Crop production
Low-input/rain-fed crop systems dominate crop production in Sub-Saharan Africa. For the 42 Sub-
Saharan African countries in the SPAM sample, 63 percent of the $53 billion average annual production
(1999–2001) of the 20 major crops is cultivated under such systems (table 2.1). The remainder of the
production is nearly evenly split between high-input/rain-fed (20 percent) and irrigated systems (16
percent).11
For the six major cereal crops (wheat, rice, maize, barley, millet, and sorghum), the overall
distribution is similar: 60 percent were under low-input/rain-fed, 25 percent under high-input/rain-fed,
and the remaining 15 percent under irrigated systems. But, when we look at three major cash crops
(cotton, coffee, and groundnuts), 40 percent were produced under the irrigated system, which enables the
highest yields among production systems. For example, table 2.2 shows that the potential yield of cash
crops in kilogram per hectare (kg/ha) under the irrigated production system is as many as 15 times higher
than that under low-input/rain-fed production. A summary of each of the 20 crops is given in table A.5.12
11
Again, 20 crops are aggregated using the medians of country-level individual crop prices.
12
See You and others (2007a) for a summary of country-level crop production.

6
Table 2.1 Crop production in Sub-Saharan Africa, average 1999–2001 estimates
Share (%)
Crop production (mn $) High-input/rain-fed Low-input/rain-fed Irrigated
Total crops (20) 52,822 20.8 63.1 16.1
Cereals (6) 12,997 25.0 59.6 15.4
Cash crops (3) 10,091 19.9 39.5 40.6
Source: Calculated from output of the IFPRI SPAM.
Table 2.2 Potential yield by production system in Sub-Saharan Africa (kg/ha)
High-input/rain-fed Low-input/rain-fed Irrigated
20 crops 107.7 9.5 142.0
6 cereals 254.3 33.6 406.3
3 cash crops 107.7 9.5 142.0
Coffee 134.4 13.5 186.7
Cotton 38.4 2.7 58.0
Groundnuts 150.2 12.3 181.4
Source: IFPRI SPAM, originally from FAO/IIASA.
Travel time, population, and agricultural production
To examine the relationship between connectivity/remoteness (as measured by travel time to the
nearest city of 100,000 people or more), population, and crop production in Sub-Saharan Africa, table 2.3
presents data sorted by travel-time decile.13
Given the spatial distribution of road infrastructure and
population in 2000, the average person in Sub-Saharan Africa lived 4.6 hours from a city of 100,000
people or more. The 10 percent of the population (214 million people) that lived closest to (and in) cities
of 100,000 people or more, however, had an estimated travel time of less than 1.7 hours. These urban
areas and their immediate surrounding areas accounted for nearly one-fourth of crop production in value
terms. In contrast, only 8.4 million people (1.6 percent of the population) lived in the decile farthest in
travel time from large cities, for whom the average travel time was nearly 25 hours.
Thus, not surprisingly, average travel times are inversely related to population size. On average, areas
with larger populations have better road networks and therefore less travel time to nearby cities. Total
crop production shows the same pattern: higher crop production is observed in the areas with larger
populations and better road networks. Likewise, per capita production shows a broadly similar pattern to
the first (mainly urban) decile. Thus, the average value of per capita production, which is only $58 per
person in the first decile, is $139 per person for deciles 2 through 6 and falls to $113.3 per person in
deciles 7 through 10 (table 2.4 and figure 2.1).14
13
Table A.6 presents this data for all deciles.
14
Note that the value of production per capita in decile 10 ($167 per person) is actually higher than that of all other
deciles, likely reflecting the very small population size of that decile and perhaps the relative dominance of
noncereal crops over cereals in these most remote areas.

7
Table 2.3 Travel time, population, and crop production in Sub-Saharan Africa
Travel time decile
Average
travel
time (hrs)1
Total population
(mns)
Total crop production
(mn $)
Share of high-input /
rain-fed production
Total crop production
/ potential
1 1.7 213.9 12,469.3 0.174 0.411
2 3.0 69.3 10,167.9 0.184 0.456
3 4.1 52.6 7,822.9 0.188 0.466
4 5.1 46.5 6,958.5 0.188 0.332
5 6.3 38.3 4,593.6 0.186 0.202
6 7.6 30.8 3,478.9 0.180 0.163
7 9.3 23.8 2,580.3 0.180 0.082
8 11.7 18.3 2,030.6 0.179 0.059
9 15.4 14.2 1,315.8 0.177 0.047
10 24.8 8.4 1,404.5 0.166 0.029
Total/average 4.6 516.1 52,822.3 0.182 0.191
Source: Own calculations.
Note: Travel time is the estimated time to the nearest city with 100,000 people or more; 10 percent of [land] area in Sub-Saharan Africa is
within 1.7 hours of a city with 100,000 people or more; 40 percent of [land] area is more than 7.6 hours from a city with 100,000 people or
more. Crop production is estimated using median prices (in US dollars) for each of the 20 crops.
Table 2.4 Travel time, population, and crop production in Sub-Saharan Africa
Population Crop production Production per capita
Travel time (mns) (%) (bn $) (%) ($ per person)
<2.5 hrs 213.9 41.4 12.5 23.6 58.3
2.5–8.4 hrs 237.5 46.0 33 62.5 139.0
> 8.4 hrs 64.7 12.5 7.3 13.9 113.3
Average 516.1 100.0 52.8 100.0 102.3
Note: Travel time (in hours) is calculated to the nearest city with 100,000 people or more.
Moreover, actual crop production as a share of potential, which averages 0.191 overall, declines
monotonically from the third to the tenth travel time deciles, from 0.466 to only 0.029 (table 2.3). Note
that in no case is the ratio of actual production to potential production greater than 0.466, suggesting that
agronomic potential production is perhaps an unrealistically high standard. Even so, the ratio of actual to
potential production in the last four travel-time deciles falls far below the maximum ratio actually
achieved, suggesting that there is enormous untapped potential for expansion of agricultural production in
the more remote areas of Sub-Saharan Africa. Inadequate supply of labor and insufficient demand (either
for self-consumption or for sale) are probably the major factors for lower output in these remote areas.
There are, however, some differences in the impacts of road connectivity across the three crop-
production systems. Farmers who adopt the high-input/rain-fed production system are likely to be more
sensitive to transport costs for intermediate inputs, such as fertilizer and pesticide, than those using low-
input/rain-fed technology. Therefore, we expect high-input/rain-fed production will be more concentrated
in the areas of well-connected road networks. Interestingly, the pattern we obtained from the sample
shows an inverted-U-shape relationship between the relative adoption of the high-input production system

8
and travel time (table 3.3 and figure 3.2). In the areas of well-connected road networks and large
populations, small-scale crop production done mainly for subsistence (and, therefore, under low-
input/rain-fed production) dominates other crop-production systems. In these areas of the first to third
deciles, reducing travel time actually decreases the relative share of high-input/rain-fed crop production.
But, after the median travel time areas, increasing travel time discourages the adoption of high-input
production. In the areas of the third and fourth deciles, the relative adoption of high-input crop production
is the highest.
The intensity of land
utilization for crop
production generally has a
similar, inverted U shape. In
the areas of large populations
and well-connected road
networks (and, therefore,
with short travel time), land
use for crop production has
to compete against highly
productive urban land use,
and increasing road networks
reduces land-use intensity for
crops. But increasing travel
time combined with
decreasing population
density eventually reduces
land-use intensity for crop
production. In summary, the
ratio of actual crop production to potential (maximum) crop production has an inverse-U-shape
relationship with travel time.
Figure 2.1 Travel time, population, and crop production in Sub-Saharan Africa
Note: Population and crop-production figures indicate percentages of the totals for Sub-Saharan
Africa. Per capita production figures indicate percentage of the average for Sub-Saharan Africa.
0%
20%
40%
60%
80%
100%
120%
140%
160%
Population Crop Production Per Capita Production
<2.5 hrs 2.5-8.4 hrs > 8.4 hrs

9
Figure 2.2 Crop production and population distribution, by travel time
0510152025
0 2 4 6 8 10
Travel time decile
Travel time to nearest city with 100,000 or more, hours
050100150200
population,mil.
0 2 4 6 8 10
Travel time decile
Population, mil.
0
50001000015000
0 2 4 6 8 10
Travel time decile
total crop production, mil $ high-input rainfed, mil $
low-input rainfed, mil $ irrigated, mil $
Total crop production, mil $
.165.17.175.18.185.19
0 2 4 6 8 10
Travel time decile
Share of high-input rainfed total crop production
0.1.2.3.4.5
0 2 4 6 8 10
Travel time decile
Ratio of total crop production over potential production
Note: Travel time deciles are based on travel time to the nearest city with 100,000 people or more.

10
Estimated crop-surplus regions
Based on the observed patterns above, we develop an indicator that gives an approximate measure of
the crop surplus or deficit in each pixel. The indicator measures the gap between crop production and
demand. The supply side is simple: the total crop value produced in a pixel. On the crop-demand side, we
measure effective (or normative) demand, rather than observed (or supply-constrained) demand. The
basic assumption is that the total crop demand in a pixel is proportional to the total population size in the
pixel. To make supply and demand comparable, we normalize crop production and demand in each pixel
using sample averages (Sub-Saharan African total).
i
i
i
i
total crop production
normalized crop production
mean of total crop production
population
normalized crop demand
mean of population
=
=
As crop production is the major source of rural household income, this crop balance indicator also
indirectly measures rural income generation relative to population. But we do not include food imports in
the analysis as in Miller
(2001).15
Figure 2.3 and table
2.5 show how crop
balance varies across
different travel time
deciles. As travel time
(to the nearest city with
100,000 people or more)
increases, crop
production responds by
decreasing almost
linearly, but crop
demand (proportional to
population) decreases
exponentially. This
pattern can be explained
by agglomeration
economies.
Agglomeration of
populations near well-
connected road networks produces positive externalities, or external economies of scale, which in turn
induce people to cluster more. This process creates urban centers and clustering of road networks where
15
Miller (2001) calculates food supply as the sum of total domestic food production and food imports. For food
demand, in scenario (i) he uses the 1998 Food and Agriculture Organization Statistical Databases (FAOSTAT)
national averages for daily caloric consumption and the 1998 Oak Ridge National Laboratory (ORNL) population
density data, and in scenario (ii) he uses an average food consumption of 2,000 calories per day per person.
Figure 2.3 Crop surplus and shortage, by travel time
Note: Travel time deciles are based on travel time to the nearest city with 100,000 people or more.
-2024
0 2 4 6 8 10
Travel time decile
crop surplus/shortage=(a)-(b) total crop production,normalized(a)
total population,normalized(b)
Crop Surplus/Shortage

11
people and urban economic activities agglomerate more densely (Henderson and others 2001). But crop
production may not exhibit the same magnitude of scale economies as urban activities, and clustering of
crop production in response to road connectivity is limited.16
The first decile area shows the largest shortage of crop production, as food demand exceeds local crop
production in areas close to and including large cities. This deficit is filled by surplus crop production
mainly in the second to fourth decile areas. Interestingly, the large deficit in the first decile is almost
cancelled out by the surplus in the next three deciles, so that, overall, the remote areas after the fifth decile
(with weaker road networks and higher travel times to large cities) are generally in an autarkic situation,
with local demand met by local crop production.
When these crop balances are aggregated to the country level, Ethiopia, Tanzania, Kenya, Sudan,
South Africa, and Zambia show significant crop deficits, while Zimbabwe, Uganda, and Côte d’Ivoire
exhibit large crop surpluses (table A.6). When we look at regional patterns, most West African countries
near the Gulf of Guinea show large crop surpluses, whereas many East African countries display
significant crop shortages, with the few exceptions of Malawi, Madagascar, Uganda, and Zimbabwe.
Table 2.5 Estimated crop surplus and deficit by travel time deciles
Decile
(number
of pixels)
Travel time
(hours)1
(a)
Crop production
normalized2
(b)
Population
normalized
(a–b)
Crop surplus
or deficit
1 (14,762) 1.7 2.36 4.15 –1.78
2 (14,763) 3.0 1.93 1.34 0.58
3 (14,762) 4.1 1.48 1.02 0.46
4 (14,763) 5.1 1.32 0.90 0.42
5 (14,763) 6.3 0.87 0.74 0.13
6 (14,762) 7.6 0.66 0.60 0.06
7 (14,763) 9.3 0.49 0.46 0.03
8 (14,762) 11.7 0.39 0.35 0.03
9 (14,763) 15.4 0.25 0.28 –0.03
10 (14,819) 24.8 0.27 0.16 0.10
Note: 1. Travel time to the nearest city with 100,000 people or more, in hours.
2. 20 crops, aggregated using the medians of country-level crop prices.
Figure 2.4 maps pixel-level crop balances in Sub-Saharan Africa. We observe significant spatial
variations in crop surplus and shortages even within a country. As crops need to be transported and traded
from crop surplus areas to deficit areas, mapping spatial distribution of crop balances could provide
important insights into potential crop trade flows and corresponding transportation and logistic demands.
16
Physical constraints, such as limited suitable land areas, also contribute to less clustering in crop production.

12
Figure 2.4 Sub-Saharan African crop balances by country and pixel
3 Conceptual framework and model
In assessing the implications of location and investments in transport costs on crop production and
productivity in Sub-Saharan Africa, we adopt a conceptual framework in which transport investments
affect both the supply and demand for crop production.
On the supply side, production of crop j under production system l in location (pixel) i depends on the
agronomic potential pj, under the production system l in location i, and unobserved location-specific
variables ( i) such as output and factor prices and available technology.
Demand for a crop produced in location i depends on the size of the local market surrounding location
i, which is in turn determined by population, distribution of per capita incomes, and the trade regime
(especially whether the domestic market is integrated with the international market).
Raster_cropzone3.shp
-299.06 - -1.06
-1.06 - -0.45
-0.45 - -0.24
-0.24 - -0.12
-0.12 - -0.05
-0.05 - -0.01
-0.01 - 0.04
0.04 - 0.36
0.36 - 1.4
1.4 - 238.76
Countries.shp

13
The hypothesis to be tested is that better road connectivity increases crop production (or productivity)
after controlling for other factors. This effect is assumed to take place through reduction in transport costs
of goods and services that raise crop producers’ prices (depending on the elasticity of demand as well as
supply).17
Reduced transport costs also lower producers’ input costs (increasing profitability and supply)
and may increase access to technology, including extension services.
There are also possible compositions of production effects, as lower transport costs result in a greater
percentage reduction in the price of perishable, bulky items such as vegetables, thus increasing their
relative profitability. Finally, where the transport cost reduction is sufficiently large and widespread, there
are potential general equilibrium effects on rural and urban nonfarm sectors, wages, and overall incomes
(Diao and others 2006).
Thus, the basic model, expressed as a reduced-form crop-production function, is:
Crop productionijl = f(agronomic potentialijl, local market sizei, road connectivityi, )
Agronomic potential, or potential crop production, reflects the supply-side constraints on the quantity
and quality of suitable land. It measures the maximum crop-production estimates, given suitable land size
and quality, in a pixel. In estimation, we multiply the suitable land size by corresponding potential yield
(of each crop and production system) in a pixel. Data for both variables are from FAO/IIASA and are
used as inputs in the IFPRI SPAM.
To control for the local market size, a main determinant of crop demand, we propose two variables:
population count in a pixel and the distance-weighted population aggregate in neighboring areas (up to a
100-km radius). Road connectivity is measured by the travel time to the nearest city with (a) 25,000
people or more, (b) 100,000 people or more, or (c) 500,000 people or more. We experiment with these
three road-connectivity measures in turn. Finally, we add country dummies to control for (unobserved)
country fixed effects.
Econometric issues
First, it is necessary to correct for bias in the regression estimates arising because the dependent
variables (crop production and productivity) are left-censored data (that is, by definition, their values are
never less than a certain value, in this case zero).18
To overcome this potential bias, we estimate the
equations using a Tobit (censored regression) model. In addition, we have dropped observations (pixels)
that are unsuitable for agricultural production from our regression.
Second, in the basic model, one of the explanatory (right-hand-side) variables—road connectivity—
can be endogenous, since it may be determined by unobserved local factors that also influence crop
17
The price gain to producers increases as demand becomes more elastic.
18
A crop is produced in an area only when the climate and soil are suitable for the crop and when its production is
profitable (the revenue generation from the crop production should be greater than the costs). In mathematical terms,
we may express this in the crop-production function: yi* = ’xi + i. We observe crop production in location i
(yi=yi* if yi*>0) only when it is profitable; otherwise, there is no production in that location (yi=0 if yi*<0).
Standard regression methods fail to account for the qualitative difference between limit (zero) observations and
nonlimit (continuous) observations and thus produce biased parameter estimates (Greene 2003). In our data set,
there are in fact many locations where production of certain crops is zero.

14
production. For example, a road may have been initially constructed primarily to connect a mining area to
a port. There may likewise be a concentration of food production around the mining center because of the
demand for food of the population working in the mining area. In this case, both crop production and road
connectivity are determined by a common exogenous factor (location of a mining area).
Moreover, measurement error (poor road data quality that influence our travel-time estimates) may
also lead to attenuation bias (biasing parameter estimates toward zero).
To minimize the magnitude of possible bias arising from the endogeneity of road connectivity and
measurement error, we use predicted values of the road connectivity indices from instrumental variable
regressions. For this first-stage regression, we use terrain roughness (difference between maximum and
minimum elevations), total populations of each cell and its surrounding areas (within 100 km), population
density, and the exogenous variables from the main regression as instruments. The Wald test of the
exogeneity of the instrumented variables in the IV-Tobit estimation confirms the endogeneity of travel-
time variables in most specifications.19
Combining the Tobit model and IV estimation, we use the IV-Tobit estimation, or the Tobit model
with endogenous regressors, for the econometric analysis.20
4 Econometric estimates
The impacts of road connectivity on crop production
Table 4.1 shows the basic results of the regressions for road-connectivity impacts on total crop
production under 3 production systems in all 42 Sub-Saharan African countries in the sample. All the
regressors are significant with expected coefficient signs. Physical constraints (potential maximum
production) and local market size have significant effects on total crop production. After controlling for
these effects and (unobserved) country fixed effects, road connectivity measures and travel times to the
nearest cities have significant effects on total crop production across three crop-production systems.
Two distinct patterns emerge from this analysis. First, the elasticity of crop production increases (in
absolute magnitude) as we measure travel time to “larger” cities. For example, under low-input
production systems, the elasticities of travel time to the nearest city with a population of 25,000, 100,000,
and 500,000 are –2.3, –2.9, and –4.8, respectively. Under high-input production systems, these elasticities
are –1.2, –1.6, and –1.6, respectively; and for irrigated systems, the corresponding elasticities are –1.5, –
1.8 and –3.5. Thus, the regressions suggest that there is much greater concentration of production in
regions surrounding large cities than in regions surrounding smaller ones. This pattern is consistent with
previous findings given in table 4.1 and figure 2.3, which show significant crop shortages in the first
decile areas (mainly urban centers) and large crop surpluses in the second to fourth decile areas.
19
In most cases, we can easily reject the null hypothesis of the exogeneity of travel-time variables.
20
Specifically, we use Newey’s efficient two-step estimator, which is easier than the conditional maximum-
likelihood estimator.

15
Table 4.1 The impact of road connectivity on crop production: Sub-Saharan Africa total
Equations (1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable
Ln (total crop production,
high-input)
low-input/subsistence)
irrigated)
Estimation method IV-Tobit IV-Tobit IV-Tobit IV-Tobit IV-Tobit IV-Tobit IV-Tobit IV-Tobit IV-Tobit
–1.189*** –2.251*** –1.479***Ln (travel time to
nearest 25K city) (0.075) (0.064) (0.089)
–1.557*** –2.864*** –1.807***Ln (travel time to
nearest 100K city) (0.090) (0.079) (0.106)
–1.593*** –4.814*** –3.508***Ln (travel time to
nearest 500K city) (0.159) (0.153) (0.207)
1.193*** 1.197*** 1.194***Ln (potential production
high-input) (0.010) (0.010) (0.010)
low-input) (0.008) (0.008) (0.008)
irrigated) (0.008) (0.008) (0.009)
0.115*** 0.119*** 0.132*** 0.159*** 0.163*** 0.192*** 0.020** 0.031*** 0.045***
Ln (population)
(0.007) (0.007) (0.006) (0.005) (0.006) (0.006) (0.008) (0.008) (0.008)
0.249*** 0.104*** 0.180*** –0.093*** –0.328*** –0.728*** –0.308*** –0.453*** –0.855***Ln (neighbor population
aggregate) (0.029) (0.035) (0.050) (0.025) (0.030) (0.046) (0.035) (0.042) (0.064)
Country dummies yes yes yes yes yes yes yes yes yes
Constant yes yes yes yes yes yes yes yes yes
Wald test of exogeneity 235.50 252.75 87.31 1,146.17 1,286.72 1,188.36 343.29 378.44 469.19
(p-value) 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Left-censored
observations
25,793 25,793 25,793 10,990 10,990 10,990 34,780 34,780 34,780
Uncensored
observations
100,189 100,189 100,189 116,677 116,677 116,677 71,742 71,742 71,742
Total observations 125,982 125,982 125,982 127,667 127,667 127,667 106,522 106,522 106,522
Note: 1. Ln (x) refers to the natural logarithm of variable x.
2. Total crop production (value) is the 20-crop aggregate using the medians of African country-level crop prices.
3. Standard errors in parentheses.
4. * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
The second broad pattern observable in these regressions is that the elasticities of travel time on crop
production under the low-input crop-production system are almost two times higher than those under
high-input and irrigated systems. For example, a 1 percent reduction in travel time to the nearest city with
100,000 people or more increases low-input crop production by 2.9 percent, but results in only a 1.6
percent increase in high-input crop production and a 1.8 percent increase in irrigated crop production. For
travel time to the nearest city with 500,000 people or more, the elasticity on low-input production is –4.8,
compared to –1.6 for high-input and –3.5 for irrigated production. For travel time to the nearest city with
25,000 people or more, the same coefficients are –2.3, –1.2, and –1.5, respectively. As discussed below,
the implied own-price elasticities of supply from these regressions are extremely high, suggesting that

16
these travel-time elasticities are capturing a long-term relationship involving the location of population
centers (markets) and location of crop output, particularly for low-input technologies. For high-input/rain-
fed and irrigated crop-production systems, which often involve large farms (for example, as in Sudan),
location of production may be driven more by the availability of irrigation facilities and water than by
self-consumption or proximity to final markets in urban centers.
Table 4.2 tests the robustness of the results across different crop categories: (a) six cereal crops of
wheat, rice, maize, barley, millet, and sorghum and (b) maize (the most commonly produced crop in
Africa). Overall, we observe the same patterns as in the regressions for all crops combined. Low-
input/rain-fed crop production has the highest elasticity of road connectivity on crop production. Better
road connection to larger cities has larger impacts on crop production.
Table 4.2 Impacts of road connectivity on the production of different crops: Sub-Saharan Africa total
(1) (2) (3) (4)
Elasticity of travel time
to nearest city with
100,000 people
High-input crop
production
Low-input and
subsistence crop
production
Irrigated crop
production All combined
Estimation method IV–Tobit IV–Tobit IV–Tobit IV–Tobit
–1.557*** –2.864*** –1.807*** –2.396***
Total crops
(0.090) (0.079) (0.106) (0.059)
— –1.975*** –2.488*** –2.003***
Cereals
— (0.099) (0.135) (0.087)
–3.649*** –3.933*** 1.862*** –3.132***
Cash crops
(0.033) (0.110) (0.136) (0.104)
— –2.829*** –1.746*** –3.331***
Maize
— (0.149) (0.156) (0.146)
Note:
1. Total crops include 20 crops; cereals (6) are wheat, rice, maize, barley, millet, and sorghum; cash crops (3) are coffee, cotton, and
groundnuts. All crops are aggregated using the medians of African country-level crop prices (US$).
2. Standard errors in parentheses.
3. * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Road connectivity and adoption of high-input/high-yield technology: Sub-
Saharan Africa total
In this section, we examine the impacts of road connectivity on the adoption of high-input/high-yield
crop-production technology. We measure the adoption of high-input technology by the share of high-
input/rain-fed production in total crop production, again for the 20-crop aggregate. For a measure of high-
yield crop-production technology, we choose high-input/rain-fed crop production rather than the irrigated
system, which is mainly constrained by external factors (irrigation infrastructure) that farmers cannot
control.
For control variables, we use the same group of regressors as in the previous section. But to control
for physical constraints, we use the share of high-input potential (maximum) production in total potential

17
production in each pixel. The total potential production (value) is the sum of potential production
(potential yields multiplied by suitable areas) across the three production systems in a pixel.
The results in table 4.3 are consistent with our prior result. Longer travel time discourages the
adoption of high-input/high-yield crop-production technology more than other production systems. Better
road connectivity makes high-input production more profitable and therefore increases its share of
production. This shift toward high-input production systems is driven through both direct and indirect
channels. In the direct channel, roads increase crop production by shifting outward both the crop demand
curve (through access to a larger market) and the crop supply curve (through better access to intermediate
inputs and new technology). In the indirect channel, roads facilitate the adoption of high-input/high-yield
crop production and therefore increase crop production by replacing low-input/low-yield crop production.
Table 4.3 Impacts of road connectivity on the adoption of high-input crop production system: Sub-Saharan Africa total
Equations (10) (11) (12)
Dependent variable Share of high-input crop production in total crop production
Estimation method IV–Tobit IV–Tobit IV–Tobit
–0.020***
Ln (travel time to nearest 25K city)
(0.005)
–0.027***
(0.006)
–0.089***
(0.011)
0.149*** 0.150*** 0.151***Share of high-input potential production in total potential
production of all crops (0.004) (0.004) (0.004)
–0.001* –0.001 –0.000
Ln (population)
(0.000) (0.000) (0.000)
0.016*** 0.013*** –0.005
Ln (neighbor population aggregate)
(0.002) (0.002) (0.003)
Country dummies yes yes yes
Constant yes yes yes
Wald test of exogeneity 25.25 25.64 71.77
(p-value) 0.000 0.000 0.000
Left-censored observations 17,596 17,596 17,596
Uncensored observations 83,129 83,129 83,129
Total observations 100,725 100,725 100,725
Note: Total crop production (value) is 20-crop aggregate using the medians of African country-level crop prices. Standard errors in
parentheses.
* significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Table 4.3 also confirms that the impacts of road connectivity on the adoption of high-input/high-yield
crop production are consistent across the different measures of travel time—to the nearest city with a
population of 25,000, 100,000, and 500,000 or more. Interestingly, the impact is higher as we consider
road connectivity to more populous cities. The semielasticity on the share of high-input crop production is

18
–0.09, –0.03, and –0.02 for travel time to the nearest city with a population of 500,000, 100,000, and
25,000 or more, respectively.
Comparing East and West Africa
The above regressions for all of Sub-Saharan Africa include countries with vast differences in
agroecological conditions, infrastructure, and crops. Although these regressions include country dummies
that are designed to capture fixed effects at the country level, we also run regressions on more
homogeneous sets of countries, namely five coastal West African countries of the humid and subhumid
tropics where root crops and cereals dominate food crop-production systems (Nigeria, Benin, Togo,
Ghana, and Côte d’Ivoire) and seven large countries of East and southern Africa (Kenya, Uganda,
Tanzania, Zambia, Malawi, Mozambique, and South Africa) where maize is the major staple.
On average, the East African region has lower population density, smaller local markets (based on the
market definition in the previous sections), and lower road connectivity (table A.7). For example, the
average population density in East Africa is just 43 percent of that in West Africa; the average population
in a pixel, 42 percent; and the distance-weighted population aggregate, 40 percent. Interestingly, the
average travel time to the nearest city with 25,000 people or more in East Africa is 242 percent of that in
West Africa; of 100,000 or more, 224 percent; and of 500,000 or more, 221 percent. This large difference
in travel time comes from type-2 (all-weather/improved) and type-3 (earth/partially improved) road
differences. The average straight line distance to type-1 (motorway/major) roads in East Africa is similar
to that in West Africa (95 percent), but the average distances to type-2 and type-3 roads in East Africa are
175 percent and 190 percent of those in West Africa, respectively.
The difference between East and West Africa is more distinct when we compare the average total
crop production per pixel in the two regions. First, there is no large difference in terms of average suitable
areas and potential (maximum) production in a pixel. The average suitable area in East Africa is 93
percent of that in West Africa, and the average potential production is about 88 percent. But the average
crop production per pixel in East Africa is just 30 percent of that in West Africa. The average high-
input/rain-fed crop production in East Africa is 35 of that in West Africa; low-input/rain-fed, 31 percent;
and irrigated, 15 percent.
Given these distinct agrodemographic differences between East and West Africa, it is not surprising
that road connectivity has different impacts in the two regions. In East Africa (table 4.4), we observe
basically the same patterns as for total Sub-Saharan Africa (table 4.1). Longer travel time decreases total
crop production. This pattern is consistent across different travel-time measures and different production
systems. Again, travel time to a larger city gives higher crop-production elasticity, and the elasticity is the
highest under the low–input/rain-fed production system. It is worthwhile to note that in all specifications,
the elasticities in East Africa are lower than those in the sum of Sub-Saharan African countries. Similarly,
the elasticities from regressions for various crop subgroups for East Africa (table 4.5) are generally lower
than those for all Sub-Saharan Africa (table 4.2).

19
Table 4.4 The impacts of road connectivity on crop production: East Africa
Equations (1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable
high-input)
irrigated)
Estimation method IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit
–0.162 –1.265*** –0.781***Ln (travel time to nearest 25K
city) (0.112) (0.113) (0.104)
–0.396*** –1.818*** –1.030***Ln (travel time to nearest
100K city) (0.146) (0.152) (0.136)
–0.458 –3.946*** –2.175***Ln (travel time to nearest
500K city) (0.311) (0.354) (0.302)
1.137*** 1.141*** 1.139***Ln (potential production high-
input) (0.020) (0.020) (0.020)
0.695*** 0.669*** 0.695***Ln (potential production low-
input) (0.023) (0.023) (0.024)
irrigated) (0.013) (0.013) (0.014)
0.169*** 0.162*** 0.165*** 0.219*** 0.201*** 0.173*** 0.135*** 0.129*** 0.118***
Ln (population)
(0.012) (0.012) (0.013) (0.012) (0.013) (0.015) (0.012) (0.012) (0.013)
0.475*** 0.399*** 0.410*** 0.306*** 0.160*** –0.235*** –0.026 –0.105* –0.362***Ln (neighbor population
aggregate) (0.048) (0.056) (0.087) (0.046) (0.054) (0.090) (0.045) (0.054) (0.089)
Wald test of exogeneity 0.32 3.33 0.41 109.36 146.81 145.59 33.82 49.90 93.96
(p-value) 0.573 0.068 0.523 0.000 0.000 0.000 0.000 0.000 0.000
Left-censored observations 8,114 8,114 8,114 6,757 6,757 6,757 7,719 7,719 7,719
Uncensored observations 27,239 27,239 27,239 30,229 30,229 30,229 23,167 23,167 23,167
Note: East Africa includes the following seven countries: Kenya, Uganda, Tanzania, Zambia, Malawi, Mozambique, and South Africa. Total
crop production (value) is 20-crop aggregate using the medians of African country-level crop prices. Standard errors in parentheses.
As for all Sub-Saharan Africa, reducing travel time significantly increases the adoption of high-
input/high-yield technology in East Africa (table 4.6). The impacts are greater when roads are connected
to larger cities. The semielasticity on share of high-input crop production is -0.07, -0.02, and -0.01 for
travel time to the nearest city with a population of 500,000, 100,000, and 25,000 or more, respectively.

20
Table 4.5 Impacts of road connectivity on the production of different crops: East Africa
(1) (2) (3) (4)
Elasticity of travel time to nearest
city with 100,000 people
High-input crop
production
Low-input and
subsistence crop
production
Irrigated crop
Estimation method IV-Tobit IV-Tobit IV-Tobit IV-Tobit
–0.396*** –1.818*** –1.030*** –1.691***
Total crops
(0.146) (0.152) (0.136) (0.086)
–0.419** –0.483*** –1.836*** –1.764***
Cereals
(0.185) (0.162) (0.198) (0.130)
— –1.824*** –1.043*** –1.219***
Cash crops
— (0.190) (0.164) (0.146)
1.939*** 1.175*** –0.808*** –0.777***
Maize
(0.406) (0.326) (0.212) (0.222)
crops include 20 crops; cereals (6) are wheat, rice, maize, barley, millet, and sorghum; cash crops (3) are coffee, cotton, and groundnuts. All
crops are aggregated using the medians of African country-level crop prices (US$). Standard errors in parentheses.
The regression results for West Africa (table 4.7), however, show markedly different patterns. First,
the coefficient estimates of travel time under high-input/rain-fed systems are positive, counter to our
expectations. Second, the coefficient estimates of travel time under low-input/rain-fed systems are
negative, as expected, but the coefficient is much smaller than that of East Africa. As in East Africa and
all Sub-Saharan Africa, travel-time measures have similar elasticities across different travel-time
measures: the coefficient estimates of travel time to the nearest city with a population of 25,000, 100,000,
and 500,000 and more are -0.9, -1.1, and -1.0, respectively. The impacts of road connectivity on irrigated
crop production in West Africa are of similar magnitude, but the impacts of road connectivity on the high-
input/high-yield technology adoption in West Africa are insignificant (table 4.8).
One possible explanation for these differences across regions is that road connectivity may show
decreasing marginal productivity in crop production: the more densely roads are connected, the smaller
the marginal benefits. Since East Africa has less road infrastructure, the (marginal) impacts of new road
construction (on crop production) can be higher than those in West Africa, which already has a relatively
well-connected road network.

21
Table 4.6 Impacts of road connectivity on the adoption of high-input crop production system: East Africa
Equation (1) (2) (3)
–0.012**
(0.006)
–0.019**
(0.008)
–0.067***
(0.016)
–0.005*** –0.005*** –0.006***
Ln (population)
(0.001) (0.001) (0.001)
0.012*** 0.009*** –0.003
(0.002) (0.003) (0.005)
(p-value) 0.176 0.330 0.067
crop production (value) is 20-crop aggregate using the medians of African country-level crop prices. Standard errors in parentheses.

22
Table 4.7 Impacts of road connectivity on crop production: West Africa
Equations (1) (2) (3) (4) (5) (6) (7) (8) (9)
Dependent variable
high-input)
irrigated)
Estimation method IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit IV–Tobit
0.700*** –0.921*** –0.693***Ln (travel time to nearest 25K
city) (0.182) (0.116) (0.219)
0.463** –1.102*** –0.573**Ln (travel time to nearest
100K city) (0.213) (0.136) (0.278)
4.056*** –1.033*** –1.952***Ln (travel time to nearest
500K city) (0.351) (0.199) (0.422)
1.725*** 1.729*** 1.578***Ln (potential production high-
input) (0.032) (0.032) (0.038)
0.420*** 0.406*** 0.440***Ln (potential production low-
input) (0.018) (0.018) (0.019)
irrigated) (0.021) (0.021) (0.022)
0.271*** 0.242*** 0.217*** –0.009 0.002 0.063*** –0.158*** –0.132*** –0.107***
Ln (population)
(0.029) (0.028) (0.028) (0.018) (0.018) (0.016) (0.035) (0.035) (0.028)
0.340*** 0.338*** 1.846*** 0.793*** 0.645*** 0.566*** 0.263*** 0.231** –0.375**Ln (neighbor population
aggregate) (0.061) (0.080) (0.152) (0.039) (0.052) (0.085) (0.065) (0.096) (0.175)
Wald test of exogeneity 18.21 1.90 171.60 44.17 47.90 14.26 9.61 4.64 23.81
(p-value) 0.000 0.168 0.000 0.000 0.000 0.000 0.002 0.031 0.000
Left-censored observations 1,831 1,831 1,831 104 104 104 3,427 3,427 3,427
Uncensored observations 13,730 13,730 13,730 15,446 15,446 15,446 9,709 9,709 9,709
Note: West Africa includes five countries: Nigeria, Benin, Togo, Ghana, and Côte d’Ivoire. Total crop production (value) is 20-crop aggregate
using the medians of African country-level crop prices. Standard errors in parentheses.

23
Table 4.8 Impacts of road connectivity on the adoption of high-input crop-production systems: West Africa
Equations (10) (11) (12)
0.013
(0.010)
0.012
(0.013)
0.040*
(0.021)
0.006*** 0.006*** 0.005***
Ln (population)
(0.002) (0.002) (0.001)
0.013*** 0.014*** 0.026***
(0.003) (0.004) (0.009)
(p-value) 0.688 0.320 0.852
Note: West Africa includes five countries: Nigeria, Benin, Togo, Ghana, and Côte d’Ivoire. Total crop production (value) is the 20-crop
aggregate using the medians of African country-level crop prices. Standard errors in parentheses.

24
5 Interpretation of results and implications for
spatial investment
The coefficients of the variables for travel time to the nearest city would seem to imply a very large
potential response of agricultural production to infrastructure investments and policy changes that reduce
travel times. For example, if marketing margins account for 50 percent of the urban price (for example,
100 shillings of a 200-shilling urban consumer price) and if 80 percent (80 shillings) of the marketing
costs are due to factors that are proportional to travel time, then a 10 percent reduction in travel time
would reduce the marketing margin by 8 percent (8 shillings in this example). If consumer prices are
unchanged (due to a perfectly price-elastic demand), then producers reap the entire benefit of the lower
transport costs and producer prices rise by 8 percent (from 100 to 108 shillings).
Assuming an elasticity of supply of 0.3, production in this hypothetical scenario would increase by
2.3 percent, and the elasticity of production with respect to travel time would be 0.023/–0.10 = –0.23
(table 5.1). But if the product faces demand constraints (simulated using an own-price elasticity of
demand of –0.4), the urban price falls by 3.6 percent and the rural price increases by only 0.8 percent. In
this scenario, production rises by only 0.2 percent, so that the elasticity of production with respect to
travel time is –0.02.
Table 5.1 Travel-time elasticities with endogenous market price effects
Es = 0.3 Es = 0.8
Elastic
demand
Inelastic
demand
Elastic
demand
Inelastic
demand
Own-price elasticity of supply 0.3 0.3 0.8 0.8
Own-price elasticity of demand –10,000.0 –0.4 –10,000.0 –0.4
Urban price (% change) 0.0 –3.6 0.0 –3.8
Rural price (% change) 8.0 0.8 8.0 0.4
Production (% change) 2.3 0.2 6.3 0.3
Travel time (% change) –10.0 –10.0 –10.0 –10.0
Elasticity of supply with respect to travel time –0.23 –0.02 –0.63 –0.03
Note: Calculations assume that initially marketing margins account for 50 percent of the urban price and that 80 percent of the marketing
margins are due to factors that are proportional to travel time, so that a 10 percent reduction in travel time reduces the marketing margin by 8
percent. Base urban consumption is assumed to be 30 percent of base total consumption.
Both of these travel-time elasticities are significantly below the estimates for various crops for all
Sub-Saharan Africa or East Africa. The estimated travel time elasticities in Sub-Saharan Africa (all
cropping systems) are –2.0 for the six-cereal aggregates and –3.3 for maize (table 4.2); the corresponding
elasticites for East Africa are smaller in absolute magnitude (–1.8 and –0.8, respectively, table 4.5), but
still much larger than the estimates using an elasticity of supply with respect to price of 0.3. To

25
approximate even the relative small magnitude of the travel time elasticity for maize in East Africa (–
0.78, table 4.5) requires a very high own-price elasticity of supply of 0.8 and perfectly elastic demand (–
0.63, table 5.1).
The regression results for choice of technology may give a truer picture of realistic short- to medium-
term elasticities of production relative to reductions in travel time, one that does not implicitly involve
increases in area cultivated (which might require movements of people or reductions in fallow times). The
regressions on the share of high-input technology suggest more modest estimates of the elasticity of
supply. For example, in East Africa, the coefficients of -0.019 on the logarithm of travel time to the
nearest city with 100,000 people or more (table 4.6), suggests that a 10 percent reduction in travel time
leads to a 0.19 percent increase in the share of high-input crop production out of total crop production.
Assuming that new technology leads to a 50 percent increase in yields, production would increase by 0.1
percent. The Elasticity of supply with respect to travel time would be 0.001/–0.10 or –0.01, within the
range of travel-time elasticities implied for an overall supply elasticity of 0.3 and alternative elasticities of
demand (as seen in table 5.1).
Note, though, that the regression coefficients of the variables for travel time to nearest cities do not
reflect elasticities of the supply or technology adoption of farmers, but elasticities of supply of locations,
which in general are not 100 percent covered with crops, and reallocation of crop production across
locations. These large travel-time elasticities capture a long-term relationship involving the location of
population centers (markets) and the location of crop output, in which marketing constraints (that is,
demand constraints) are implicit and important. Indeed, in the absence of a marketing constraint, very
large elasticities of supply from investment in roads that open up new areas of production are possible
(for example, consider the “vent for surplus” models of development where introduction of new export
crops, in part made possible by investments and the opening up of the marketing chain, allowed massive
expansion of cocoa and coffee production in West Africa). In the long run, crop production can readily be
increased in many regions through increases in area cultivated.
But aggregate marketing constraints are extremely important for African agricultural products, since
demand for food crops is price inelastic. With abundant suitable area for production, but facing a limited
market, areas closer to markets do have a sizeable advantage as possible sites for production than do
distant areas. When reductions in travel time bring large additional supplies into the market, market prices
are likely to fall, dampening the incentives for production and possibly resulting in a reduction in output
in areas for which travel time did not decrease. Thus, the aggregate supply response (across all regions) is
likely to be less than implied by the results for individual locations. The econometric results for individual
locations cannot simply be aggregated to derive elasticities of supply for wider regions.21
21
There are other reasons why the estimated production/travel-time elasticities may overstate the supply response.
Because much of production is consumption, location of production and population tend to be spatially correlated,
even though in the short run there is no direct causality between travel times and production. Moreover, in the short
run, other constraints on production may be binding, including price distortions arising from trade and other
country-level policies and insufficient availability of labor (especially during peak labor demand seasons), fertilizer,
and credit.

26
Implications of investment: corridors versus rural roads in Mozambique
The above approach can be used to estimate the effects of alternative road infrastructure investments
on agricultural production in Mozambique, a country with one of the most detailed road infrastructure
spatial data sets in Sub-Saharan Africa. Rather than using the estimated parameters from East Africa or all
of Sub-Saharan Africa, we conduct a separate regression analysis for Mozambique and then use the
parameters to estimate the impact on agricultural productivity due to changes in travel time brought about
by alternative road investments.22
The regression results indicate considerably greater sensitivity of production to differences in travel
times in Mozambique than in East Africa overall. For example, the elasticity of total crop production with
respect to travel time to cities with 100,000 people or more in East Africa overall is - 1.7 (and -0.8 for
maize). For Mozambique, the elasiticity of total crop production with respect to travel time to cities of
50,000 or more is -2.8 for all crops (-1.6 for maize) (see tables 5.2 and 5.3).23
We use these parameters to analyze two different scenarios: (a) an investment in the five major
international corridor roads in northern Mozambique and (b) in addition to the investment in corridors, the
upgrading of all existing rural and feeder road networks to gravel, good-condition surfaces, raising speed
attainable on these roads to 60 km/hr.
Using the estimated regression coefficients and the simulated changes in travel times to cities of
50,000 people or more per pixel, improvements in the national corridor raise total crop production by 24
percent and maize production by 33 percent (table 5.4). The investments in rural feeder roads in scenario
2 raise national crop production by a further 131 percent and maize production by a further 146 percent.
These simulated production gains are extremely high, reflecting (a) the huge potential for increased
production (as reflected in the difference between actual and potential production); (b) the reduced-form
estimation procedure used, which provides measures of the elasticity of crop-supply locations (including
reallocation of crop production across locations); and (c) the underlying assumption of completely price-
elastic demand (that is, that expansions in crop production and sales do not lead to a fall in crop prices).
This latter assumption is particularly unrealistic for domestic staple food crops, although a high elasticity
may be appropriate for some exported agricultural products. Nonetheless, this analysis, done for more
microscale investments where increases in production would not greatly affect the total market supply,
could suggest the potential gains of alternative road investments on agricultural productivity and farm
incomes.
22
Note that because large-scale flooding severely damaged crops in Mozambique, the IFPRI SPAM used regional
production figures from 2006 instead of 2000 as an input into the calculation of spatial crop allocations.
23
We use a smaller city-size cutoff (50,000 as opposed to 100,000) for Mozambique to reflect cities in which major
wholesale markets are located. Consistent data on cities of 50,000 to 100,000 are not available for all of Sub-
Saharan Africa.

27
Table 5.2 Impacts of road connectivity on crop production: Mozambique
Equations (1) (2) (3) (4)
Dependent variable
Ln (total crop
production, high-
input)
Ln (total crop
production, low-
input/subsistence)
Ln (total crop
production, irrigated)
Share of high-input
crop production in
total crop production
Estimation method IV-Tobit IV-Tobit IV-Tobit IV-Tobit
–2.269*** –2.728*** –2.147*** 0.009
(0.191) (0.181) (0.253) (0.008)
0.323***
Ln (potential production, high-input)
(0.049)
–0.240***
Ln (potential production, low-input)
(0.055)
0.440***
Ln (potential production, irrigated)
(0.037)
0.035***Share of high-input potential production in
total potential production of all crops (0.013)
0.202*** –0.085** 0.242*** 0.017***
Ln (population)
(0.044) (0.042) (0.055) (0.002)
–0.525*** –0.442*** 0.058 –0.007*
Ln (neighbor population, aggregate)
(0.084) (0.079) (0.115) (0.003)
Constant yes yes yes yes
Wald test of exogeneity 152.85 200.43 18.10 0.66
(p-value) 0.000 0.000 0.000 0.418
Left-censored observations 839 440 1,955 399
Uncensored observations 5,449 5,864 3,867 5,057
Total observations 6,288 6,304 5,822 5,456
Note: Total crop production (value) is a 20-crop aggregate using the medians of African country-level crop prices. Standard errors in
parentheses.
The benefits of improved road networks, however, extend past simply increasing agricultural
production. These investments also spur rural nonfarm activity and perhaps, more importantly, reduce
rural remoteness, increasing access of households to health, education, and other opportunities. In this
case, the investment in the main corridor roads significantly reduces travel time between cities on the
corridors but does not greatly affect access to markets by northern Mozambican households (table 5.5 and
figure 5.1).24
Rural remoteness in northern Mozambique (defined as greater than 5 hours travel time to a
city of at least 50,000 people) is dramatically reduced, though, in the second scenario (upgrading all
existing roads to gravel, good-condition surfaces). This investment would reduce the number of people in
northern Mozambique living in remote areas from the current 3.5 million to 0.8 million.
24
For details of the calculations, see Dorosh and Schmidt (2008).

28
Table 5.3 Impacts of road connectivity on the production of different crops: Mozambique
(1) (2) (3) (4)
Elasticity of travel time to nearest
city with 50,000 people
High-input crop
production
Low-input and
subsistence crop
production
Irrigated crop
Estimation method IV–Tobit IV–Tobit IV–Tobit IV–Tobit
–2.269*** –2.728*** –2.147*** –2.762***
Total crops
(0.191) (0.181) (0.253) (0.185)
–2.456*** –2.262*** –1.629*** –2.062***
Cereals
(0.179) (0.198) (0.220) (0.174)
–2.916*** –2.341*** – –2.531***
Cash crops
(0.611) (0.244) – (0.270)
– –2.527*** –1.042*** –1.552***
Maize
– (0.354) (0.205) (0.203)
Note: Total crops include 20 crops; cereals (6) are wheat, rice, maize, barley, millet, and sorghum; cash crops (3) are coffee, cotton, and
groundnuts. All crops are aggregated using the medians of African country-level crop prices (US$). Standard errors in parentheses.
Table 5.4 Simulated impacts of road improvement on crop production: Mozambique
(a)
Baseline (mn $)
(b)
Scenario 1:
Improving northern
corridors to
80km/hr (mn $)
(c)
Scenario 2:
Improving all rural
roads to 60km/hr
(mn $)
(b) vs (a)
% change
(c) vs (a)
% change
Total crops
All systems 1,506 1,869 3,486 24 131
High-input 169 204 356 20 110
Low-input and subsistence 1,246 1,549 2,896 24 132
Irrigated 91 104 176 15 95
Cereals 338 441 886 31 162
Cash crops 266 364 708 37 167
Maize 222 295 546 33 146
Note:
1. Baseline values are computed from the estimation results of tables 5.2 and 5.3. Specifically, ( )ˆˆ exp lni i
i i
y X= . We select
only statistically significant parameter estimates in computation. The fitted values are then normalized to make the national total the same as
the actual, such that ˆ ˆ, /i i i
i i i
baseline y y y= .
2. Scenario 1 and 2 values are computed in the same way as the baseline after substituting corresponding improved travel times in each pixel.

29
Table 5.5 Estimated costs of road network improvements in Mozambique
Additional
improved roads
(kms)
Total population
north remote
(a)
Marginal cost
(mn $)
(b)
Change in
remoteness
(mn people)
(c) = (b)/(a)
(‘000 people
per mn $)
Current — 3.50 — — —
Corridor improvement only 2,206 2.79 314 0.71 2.27
Corridor improvement and rural roads 16,800 0.81 672 1.98 2.95
Note: Remoteness defined as greater than 5 hours of travel time to a city of 50,000 people or more. Assumes cost of $40,000/km to upgrade a
one-lane gravel road to all-weather.
Figure 5.1 Effects of comprehensive rural road upgrade on estimated travel time
Estimated travel time: Corridor upgrade Estimated travel time: Complete rural road upgrade
Source: Own calculations..
But the estimated financial costs of these road investments are substantial. Based on average road-
improvement costs for various types of roads in Sub-Saharan Africa developed by the Africa
Infrastructure Country Diagnostic, the estimated cost of improving the 2,206 km of road necessary to
complete the major national and international corridors is $314 million. Upgrading 16,800 km of existing
rural roads in Mozambique (mainly in the north) to a travel speed of 60 km/hr (the speed for a gravel road
of very good quality) would require an estimated $672 million—more than twice the cost of corridor

30
upgrading but about 30 percent more effective in reducing remoteness in terms of the marginal reduction
in remoteness per million dollars invested.25
Thus, even though reducing the time distance to major
markets is critical for development, resources are limited and prioritization is necessary.
Finally, it should be noted that corridors and rural roads may affect producer incentives more by
connecting rural roads to ports than to cities of a given size. Likewise, the relevant city size for
calculation of travel times may vary by crop or even by country or region depending on market structures.
There may also be other important impacts of road connectivity in the medium term that are not well
captured in this analysis, including potential effects on rural-urban and rural-rural migration. In addition,
the impacts of corridor investments will be broader and more diverse and could include increased
productivity in urban areas. Further systematic research on these issues is needed for definitive
conclusions.
6 Summary and policy implications
Agricultural production and proximity (as measured by travel time) to urban markets are highly
correlated in Sub-Saharan Africa, even after taking agroecology into account. Likewise, adoption of high-
input technology is negatively correlated with travel time to urban centers, although adoption rates are
low throughout most of the subcontinent. The correlations between the location of population centers and
road infrastructure (as reflected in travel time) and the location of production suggest a long-run
relationship in which land is typically not a binding constraint on aggregate production but in which
demand constraints that vary over space are important.
In terms of agronomic potential, there is substantial scope for increasing agricultural production in
Sub-Saharan Africa, particularly in more remote areas. Total crop production relative to potential
production is approximately 45 percent for areas within 4 hours travel time from a city of 100,000 people.
In contrast, total crop production relative to agronomic potential is only about 5 percent for areas more
than 8 hours travel time from a city of 100,000 people. These differences in actual versus potential
production arise mainly because of the relatively small share of land cultivated out of total arable land in
more remote areas.
For these remote regions, demand constraints in terms of low population densities and large travel
times to urban centers sharply constrain production. (Low population densities also limit local labor
availability.) Reducing transport costs (travel time) to these areas would expand the feasible market size
for these regions, easing the demand constraint on production. To the extent that the expansion in
production from these areas is small in terms of the relevant regional, national, or subnational market,
average market prices outside the formerly remote region would be unaffected and significant aggregate
production increases could result. For large remote regions, however, production increases arising from
improved connectivity would potentially lead to lower average market prices and reduced production in
already-connected areas.
25
This calculation assumes a cost of $40,000/km to upgrade a one-lane gravel road to an all-weather road, reflecting
actual road costs in South Africa and other southern African countries.

31
More important, however, is the possibility that improved connectivity will not only affect access to
markets, but in the medium term lead to increased migration from remote areas to areas near urban
centers. In this scenario, local demand in the remote region decreases and production in the region may
actually fall as a result of reduced travel times. Even so, average per capita incomes could rise in the
remote region. Thus, it is crucial to evaluate not only impacts on agricultural costs and potential markets
but impacts on the broader rural economy and household behavior.

32
References
Center for International Earth Science Information Network (CIESIN), International Food Policy
Research Institute (IFPRI), and World Resources Institute (WRI). 2000. Gridded Population of
the World (GPW), Version 2. Palisades, NY: CIESIN, Columbia University.
http://sedac.ciesin.columbia.edu/plue/gpw.
Chamberlin, Jordan, Muluget Tadesse, Todd Benson, and Samia Zakaria. 2007. “An Atlas of the
Ethiopian Rural Economy: Expanding the Range of Available Information for Development
Planning.” Information Development 23 (2–3): 181–92.
Deichmann, U. 1997. “Accessibility Indicators in GIS.” Mimeo. New York: Department of Economic and
Social Information and Policy Analysis, Statistics Division.
Diao Xinshen, Peter Hazell, Danielle Resnick and James Thurlow. 2006. “The Role of Agriculture in
Development: Implications for Sub-Saharan Africa.” DGSD Discussion Paper, No. 29. IFPRI.
Washington D.C.
Dorosh, Paul, and Emily M. Schmidt. 2008. “Mozambique Corridors: Implications of Investments in
Feeder Roads.” Spatial and Local Development Team (processed), World Bank, Washington,
D.C.
Environmental Systems Research Institute (ESRI). 1993. “Digital Chart of the World.” Vector Map
(VMap) Level 0 Database 1:1,000,000-Scale Global Vector Basemap.
Fan, Shenggen, and Peter Hazell. 2001. “Returns to Public Investments in the Less-Favored Areas of
India and China.” American Journal of Agricultural Economics 83 (5): 1217–22.
Food and Agriculture Organization (FAO). 2003. World Agriculture Towards 2015/2030: An FAO
Perspective. Rome: FAO and EarthScan.
________. 1981. Report of the Agro-Ecological Zones Project. World Soil Resources Report 48 (1–4).
Rome: FAO.
FAO, IFPRI, SAGE. 2006. Agro-Maps. A Global Spatial Database of Agricultural Land-Use Statistics
Aggregated by Subnational Administrative Districts.
http://www.fao.org/landandwater/agll/agromaps/interactive/index.jsp.
Fischer, Gunther, M. Shah, H. Velthuizen, and F. Nachtergaele. 2001. Global Agro-ecological
Assessment for Agriculture in the 21st Century. Laxenburg, Austria: International Institute for
Applied Systems Analysis (IIASA).
Greene, W. 2003. Econometric Analysis. Fifth Edition. Upper Saddle River, New Jersey: Pearson
Education, Inc.
Henderson, J. V., Z. Shalizi, and A. J. Venables. 2001. “Geography and Development.” Journal of
Economic Geography 1: 81–105.

33
Miller, J. 2001. “Using GIS for Global Food and Water Balance Process Modeling.” Proceedings of the
ESRI International User Conference 2001, (On-line).
http://www.esri.com/library/userconf/proc01/professional/papers/pap556/p556.htm.
Murray, S. 2007. “Africa Infrastructure Country Diagnostic Inception Report: Creation of a GIS Database
to Support AICD Study.” Mimeo. Washington, D.C.: World Bank.
Nelson, Andy. November 2007. “Global 1 km Accessibility (Cost Distance) Model Using Publicly
Available Data.” Mimeo. Washington, D.C.: World Bank.
———. “United Nations Environment Programme (UNEP) Road Map.” Africa Population Database, IV
Version. UNEP.
Siebert, Stefan, P. Döll, and J. Hoogeveen. 2001. Global Map of Irrigated Areas Version 2.0. Germany:
Center for Environmental Systems Research, University of Kassel/Rome, Italy: FAO.
Stifel, David, and Bart Minten. 2008. “Isolation and Agricultural Productivity.” Agricultural Economics
39 (1): 1–15.
Thomas, T. 2007. “Notes on Making Africa Access Map.” Mimeo. Washington, D.C.: World Bank.
You, L., and S. Wood. 2006. “An Entropy Approach to Spatial Disaggregation of Agricultural
Production.” Agricultural Systems 90 (1–3): 329–47.
You, L., S. Wood, and U. Wood-Sichra. 2007a. “Generating Plausible Crop Distribution and Performance
Maps for Sub-Saharan Africa Using a Spatially Disaggregated Data Fusion and Optimization
Approach.” IFPRI Discussion Paper 725, IFPRI, Washington, D.C.
You, L., S. Wood, U. Wood-Sichra, and J. Chamberlin. 2007b. “Generating Plausible Crop Distribution
Maps for Sub-Saharan Africa Using a Spatial Allocation Model.” Information Development 23
(2/3): 151–9.

34
Appendixes
Table A.1 Countries in the sample
Country No. pixels Country No. pixels
Angola 13,246 Lesotho 410
Benin 1,214 Liberia 1,121
Botswana 1,720 Madagascar 3,371
Burkina Faso 2,706 Malawi 1,031
Burundi 292 Mali 3,779
Cameroon 2,550 Mauritania 1,004
Central African Republic 2,835 Mozambique 7,309
Chad 5,303 Namibia 3,045
Congo 119 Niger 2,474
Congo, Dem. Rep. of 11,015 Nigeria 9,970
Côte d’Ivoire 2,434 Rwanda 296
Djibouti 159 Senegal 1,410
Equatorial Guinea 131 Sierra Leone 804
Eritrea 301 South Africa 9,094
Ethiopia 10,391 Sudan 16,217
Gabon 600 Swaziland 220
Gambia 121 Tanzania 10,064
Ghana 1,684 Togo 681
Guinea 2,492 Uganda 1,854
Guinea Bissau 399 Zambia 7,015
Kenya 2,283 Zimbabwe 4,518
Source: Calculated from output of the IFPRI SPAM.
Table A.2 African crop prices: the medians of country-level crop prices in year 2000
Crop $/ton Crop $/ton
Wheat 157.6 Soybean 213.7
Rice 260.5 Dry beans 335.8
Maize 140.6 Other pulses 262.8
Barley 169.9 Sugarcane 16.9
Millet 156.3 Sugar beets 38.3
Sorghum 185.6 Coffee 1,190.7
Potato 221.5 Cotton 954.8
Sweet potato 157.3 Other fibers 458.6
Cassava 87.3 Groundnuts 402.3
Plantain and banana 258.9 Other oil crops 504.5
Source: IFPRI, calculated from FAO price data.

35
Table A.3 Road type and total road length in Africa
Total lengthRoad type
km Share (%)
Estimated
travel speed, km/hr
1 Motorway/major road 132,000 11 50
2 All weather/improved 282,000 22 35
3 Partially improved/earth roads 839,000 67 25
Source: Calculated from UNEP data.
Table A.4 Decay function and distance-weighted population aggregate to define local market size
Radius, km Average distance, km
(a)
Distance weight
( )0.3
1
ik a
w =
Distance-weighted population
at location k
(base: 1,000,000)
1–2 1.5 0.885 885,467
2–5 3.5 0.687 686,720
5–10 7.5 0.546 546,363
10–20 15 0.444 443,785
20–50 35 0.344 344,175
50–100 75 0.274 273,830
Note: local market sizei ik k
k
w pop=

36
Table A.5 Value of crop production by farming system
Share (%)
Crop production (mn $) High-input rain-fed Low-input rain-fed Irrigated
Wheat 776.0 41.8 29.3 28.9
Rice 2,726.0 9.8 39.7 50.6
Maize 4,980.9 35.0 59.3 5.7
Barley 165.4 0.0 99.2 0.8
Millet 1,666.8 23.4 76.3 0.2
Sorghum 2,681.4 19.4 76.5 4.1
Potato 1,261.3 34.8 64.2 1.0
Sweet potato 6,547.4 16.3 58.2 25.5
Cassava 7,075.0 13.0 87.0 0.0
Plantain and banana 7,449.4 20.7 79.0 0.3
Soybean 279.2 29.0 68.2 2.8
Dry beans 770.3 48.6 50.5 0.9
Other pulses 902.8 19.8 79.9 0.3
Sugarcane 1,016.4 14.5 19.3 66.2
Coffee 1,189.6 35.0 64.3 0.7
Cotton 6,039.5 17.3 21.9 60.8
Other fibers 47.6 0.0 100.0 0.0
Groundnuts 2,862.0 19.0 66.3 14.7
Other oil crops 4,385.6 22.9 77.1 0.0
Source: IFPRI SPAM.

37
Table A.6 Crop surplus and shortage, country aggregates
Country
Sum of crop
surplus or
shortage1 Country
Sum of crop
surplus or
shortage1 Country
Sum of crop
surplus or
shortage1
Large shortage Balanced Moderate surplus
Ethiopia –12,657.8 Mali –171.3 Burundi 208.0
Tanzania –3,330.7 Djibouti –131.1 Gabon 212.1
Kenya –3,316.8 Gambia –80.9 Cameroon 318.0
Sudan –3,035.8 Guinea Bissau –74.2 Madagascar 464.5
South Africa –3,017.1 Swaziland –21.8 Benin 1,414.9
Zambia –1,638.8 Guinea –17.4 Malawi 1,528.4
Chad –6.3
Moderate shortage Mauritania –4.6 Large surplus
Angola –1,139.4 Congo, Dem. Rep. of 1.4 Ghana 2,024.0
Burkina Faso –966.1 Equatorial Guinea 11.5 Nigeria 2,445.0
Mozambique –896.4 Central African Republic 18.1 Rwanda 2,462.6
Niger –795.5 Togo 100.5 Côte d’Ivoire 5,855.0
Sierra Leone –579.1 Senegal 132.4 Uganda 7,122.3
Liberia –353.4 Congo 142.8 Zimbabwe 8,951.1
Eritrea –337.9
Lesotho –336.7
Botswana –287.3
Namibia –247.5
Note: Country-level aggregates of crop surplus/shortage in each pixel. Crop surplus/shortage in each pixel is defined as the difference between
crop production (normalized by the average crop production in Sub-Saharan Africa) and population (normalized by the average population in
Sub-Saharan Africa).

Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis

Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis

Recommended

Recommended

More Related Content

Viewers also liked

Viewers also liked (13)

Similar to Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis

Similar to Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis (20)

Recently uploaded

Recently uploaded (20)

Crop Production and Road Connectivity in Sub-Saharan Africa: A Spacial Analysis